========================================================================
LEAKAGE-FREE ACSIncome AUDIT (disjoint target-cal/test, matched N_CAL=250)
========================================================================

Accuracy: in-domain 0.801 [0.782,0.821]  shifted 0.747 [0.736,0.756]

Transfer gap (source-fit - target-fit ECE, on target-test), per family:
  temperature  in-domain=0.0028[0.0002,0.0054]  shifted=0.0252[0.0201,0.0288]
  platt        in-domain=0.0035[-0.0013,0.0087]  shifted=0.0442[0.0336,0.0527]
  isotonic     in-domain=0.0011[-0.0019,0.0042]  shifted=0.0779[0.0676,0.0883]

========================================================================
REGIME SPLIT: mainland-only vs Puerto-Rico-involving
========================================================================
  temperature  in=0.0028  mainland=0.0026[-0.0019,0.0070]  PR=0.0590[0.0518,0.0663]
  platt        in=0.0035  mainland=0.0033[-0.0026,0.0084]  PR=0.1056[0.0876,0.1196]
  isotonic     in=0.0011  mainland=0.0235[0.0184,0.0288]  PR=0.1594[0.1399,0.1769]
  mainland:  |dT|=0.177  acc_drop=0.0049  conf_JS=0.0130
  PR-invol.: |dT|=1.059  acc_drop=0.1197  conf_JS=0.0172

========================================================================
CORE RELATIONSHIP (Spearman over shifted pairs)
========================================================================
  |T_src-T_tgt| -> gap             rho=+0.622 (p=4.98e-12)
  conf_JS       -> gap             rho=+0.279 (p=4.98e-03)
  conf_JS       -> |T_src-T_tgt|   rho=+0.310 (p=1.67e-03)
  ATC_err       -> gap             rho=+0.516 (p=3.93e-08)